#ifndef AVCODEC_X86_H264_I386_H
#define AVCODEC_X86_H264_I386_H
+#include <stddef.h>
+
#include "libavcodec/cabac.h"
+#include "cabac.h"
//FIXME use some macros to avoid duplicating get_cabac (cannot be done yet
//as that would make optimization work hard)
-#if ARCH_X86 && HAVE_7REGS && HAVE_EBX_AVAILABLE && !defined(BROKEN_RELOCATIONS)
+#if ARCH_X86 && HAVE_7REGS && !defined(BROKEN_RELOCATIONS)
static int decode_significance_x86(CABACContext *c, int max_coeff,
uint8_t *significant_coeff_ctx_base,
- int *index){
+ int *index, x86_reg last_off){
void *end= significant_coeff_ctx_base + max_coeff - 1;
- int minusstart= -(int)significant_coeff_ctx_base;
- int minusindex= 4-(int)index;
- int coeff_count;
+ int minusstart= -(intptr_t)significant_coeff_ctx_base;
+ int minusindex= 4-(intptr_t)index;
+ int bit;
+ x86_reg coeff_count;
+ int low;
+ int range;
__asm__ volatile(
- "movl "RANGE "(%3), %%esi \n\t"
- "movl "LOW "(%3), %%ebx \n\t"
+ "movl %a11(%6), %5 \n\t"
+ "movl %a12(%6), %3 \n\t"
"2: \n\t"
- BRANCHLESS_GET_CABAC("%%edx", "%3", "(%1)", "%%ebx",
- "%%bx", "%%esi", "%%eax", "%%al")
+ BRANCHLESS_GET_CABAC("%4", "%6", "(%1)", "%3",
+ "%w3", "%5", "%k0", "%b0", "%a13")
- "test $1, %%edx \n\t"
+ "test $1, %4 \n\t"
" jz 3f \n\t"
+ "add %10, %1 \n\t"
- BRANCHLESS_GET_CABAC("%%edx", "%3", "61(%1)", "%%ebx",
- "%%bx", "%%esi", "%%eax", "%%al")
+ BRANCHLESS_GET_CABAC("%4", "%6", "(%1)", "%3",
+ "%w3", "%5", "%k0", "%b0", "%a13")
- "mov %2, %%"REG_a" \n\t"
- "movl %4, %%ecx \n\t"
+ "sub %10, %1 \n\t"
+ "mov %2, %0 \n\t"
+ "movl %7, %%ecx \n\t"
"add %1, %%"REG_c" \n\t"
- "movl %%ecx, (%%"REG_a") \n\t"
+ "movl %%ecx, (%0) \n\t"
- "test $1, %%edx \n\t"
+ "test $1, %4 \n\t"
" jnz 4f \n\t"
- "add $4, %%"REG_a" \n\t"
- "mov %%"REG_a", %2 \n\t"
+ "add $4, %0 \n\t"
+ "mov %0, %2 \n\t"
"3: \n\t"
"add $1, %1 \n\t"
- "cmp %5, %1 \n\t"
+ "cmp %8, %1 \n\t"
" jb 2b \n\t"
- "mov %2, %%"REG_a" \n\t"
- "movl %4, %%ecx \n\t"
+ "mov %2, %0 \n\t"
+ "movl %7, %%ecx \n\t"
"add %1, %%"REG_c" \n\t"
- "movl %%ecx, (%%"REG_a") \n\t"
+ "movl %%ecx, (%0) \n\t"
"4: \n\t"
- "add %6, %%eax \n\t"
- "shr $2, %%eax \n\t"
-
- "movl %%esi, "RANGE "(%3) \n\t"
- "movl %%ebx, "LOW "(%3) \n\t"
- :"=&a"(coeff_count), "+r"(significant_coeff_ctx_base), "+m"(index)
- :"r"(c), "m"(minusstart), "m"(end), "m"(minusindex)
- : "%"REG_c, "%ebx", "%edx", "%esi", "memory"
+ "add %9, %k0 \n\t"
+ "shr $2, %k0 \n\t"
+
+ "movl %5, %a11(%6) \n\t"
+ "movl %3, %a12(%6) \n\t"
+ :"=&q"(coeff_count), "+r"(significant_coeff_ctx_base), "+m"(index),
+ "=&r"(low), "=&r"(bit), "=&r"(range)
+ :"r"(c), "m"(minusstart), "m"(end), "m"(minusindex), "m"(last_off),
+ "i"(offsetof(CABACContext, range)), "i"(offsetof(CABACContext, low)),
+ "i"(offsetof(CABACContext, bytestream))
+ : "%"REG_c, "memory"
);
return coeff_count;
}
static int decode_significance_8x8_x86(CABACContext *c,
uint8_t *significant_coeff_ctx_base,
- int *index, const uint8_t *sig_off){
- int minusindex= 4-(int)index;
- int coeff_count;
+ int *index, x86_reg last_off, const uint8_t *sig_off){
+ int minusindex= 4-(intptr_t)index;
+ int bit;
+ x86_reg coeff_count;
+ int low;
+ int range;
x86_reg last=0;
+ x86_reg state;
__asm__ volatile(
- "movl "RANGE "(%3), %%esi \n\t"
- "movl "LOW "(%3), %%ebx \n\t"
+ "movl %a12(%7), %5 \n\t"
+ "movl %a13(%7), %3 \n\t"
- "mov %1, %%"REG_D" \n\t"
+ "mov %1, %6 \n\t"
"2: \n\t"
- "mov %6, %%"REG_a" \n\t"
- "movzbl (%%"REG_a", %%"REG_D"), %%edi \n\t"
- "add %5, %%"REG_D" \n\t"
+ "mov %10, %0 \n\t"
+ "movzbl (%0, %6), %k6 \n\t"
+ "add %9, %6 \n\t"
- BRANCHLESS_GET_CABAC("%%edx", "%3", "(%%"REG_D")", "%%ebx",
- "%%bx", "%%esi", "%%eax", "%%al")
+ BRANCHLESS_GET_CABAC("%4", "%7", "(%6)", "%3",
+ "%w3", "%5", "%k0", "%b0", "%a14")
- "mov %1, %%edi \n\t"
- "test $1, %%edx \n\t"
+ "mov %1, %k6 \n\t"
+ "test $1, %4 \n\t"
" jz 3f \n\t"
- "movzbl "MANGLE(last_coeff_flag_offset_8x8)"(%%edi), %%edi\n\t"
- "add %5, %%"REG_D" \n\t"
+ "movzbl "MANGLE(last_coeff_flag_offset_8x8)"(%k6), %k6\n\t"
+ "add %9, %6 \n\t"
+ "add %11, %6 \n\t"
- BRANCHLESS_GET_CABAC("%%edx", "%3", "15(%%"REG_D")", "%%ebx",
- "%%bx", "%%esi", "%%eax", "%%al")
+ BRANCHLESS_GET_CABAC("%4", "%7", "(%6)", "%3",
+ "%w3", "%5", "%k0", "%b0", "%a14")
- "mov %2, %%"REG_a" \n\t"
- "mov %1, %%edi \n\t"
- "movl %%edi, (%%"REG_a") \n\t"
+ "mov %2, %0 \n\t"
+ "mov %1, %k6 \n\t"
+ "movl %k6, (%0) \n\t"
- "test $1, %%edx \n\t"
+ "test $1, %4 \n\t"
" jnz 4f \n\t"
- "add $4, %%"REG_a" \n\t"
- "mov %%"REG_a", %2 \n\t"
+ "add $4, %0 \n\t"
+ "mov %0, %2 \n\t"
"3: \n\t"
- "addl $1, %%edi \n\t"
- "mov %%edi, %1 \n\t"
- "cmpl $63, %%edi \n\t"
+ "addl $1, %k6 \n\t"
+ "mov %k6, %1 \n\t"
+ "cmpl $63, %k6 \n\t"
" jb 2b \n\t"
- "mov %2, %%"REG_a" \n\t"
- "movl %%edi, (%%"REG_a") \n\t"
+ "mov %2, %0 \n\t"
+ "movl %k6, (%0) \n\t"
"4: \n\t"
- "addl %4, %%eax \n\t"
- "shr $2, %%eax \n\t"
-
- "movl %%esi, "RANGE "(%3) \n\t"
- "movl %%ebx, "LOW "(%3) \n\t"
- :"=&a"(coeff_count),"+m"(last), "+m"(index)
- :"r"(c), "m"(minusindex), "m"(significant_coeff_ctx_base), "m"(sig_off)
- : "%"REG_c, "%ebx", "%edx", "%esi", "%"REG_D, "memory"
+ "addl %8, %k0 \n\t"
+ "shr $2, %k0 \n\t"
+
+ "movl %5, %a12(%7) \n\t"
+ "movl %3, %a13(%7) \n\t"
+ :"=&q"(coeff_count),"+m"(last), "+m"(index), "=&r"(low), "=&r"(bit),
+ "=&r"(range), "=&r"(state)
+ :"r"(c), "m"(minusindex), "m"(significant_coeff_ctx_base), "m"(sig_off), "m"(last_off),
+ "i"(offsetof(CABACContext, range)), "i"(offsetof(CABACContext, low)),
+ "i"(offsetof(CABACContext, bytestream))
+ : "%"REG_c, "memory"
);
return coeff_count;
}
-#endif /* ARCH_X86 && HAVE_7REGS && HAVE_EBX_AVAILABLE */
- /* !defined(BROKEN_RELOCATIONS) */
+#endif /* ARCH_X86 && HAVE_7REGS && !defined(BROKEN_RELOCATIONS) */
#endif /* AVCODEC_X86_H264_I386_H */